A novel hybrid machine learning approach for change detection in remote sensing images

Change detection can play an essential role in satellite surveillance. With the availability of satellite images of a certain geographical area captured in different time instances, change detection is considered a tough task in the field of satellite applications. This research proposes a novel hybrid machine learning change detection technique from satellite images. The proposed hybrid learning approach is designed based on supervised and unsupervised learning techniques that considers the local association of adjacent pixels of the satellite images. Hybridization of clustering, soft labeling using fuzzy logic, Support Vector Machine (SVM) and Genetic Algorithm (GA) are used in change detection. Radial Basis Function (RBF) is used as the kernel function in SVM, and the RBF kernel parameters such as C and σ are optimized using GA for additional improvement of the performance. To demonstrate the efficiency of the approach, tests are performed on two satellite images captured in two different time instances on a particular geographical area. Change detection accuracy is used to validate the performance. Outcomes are compared with existing approaches and found to be superior.